What hypermedia was supposed to be — Bush, the Memex, and the founding claim

The founding document of hypermedia is an essay published in July 1945, three months before the war it was a response to had formally ended. The essay was Vannevar Bush’s “As We May Think,” and it appeared in The Atlantic Monthly under a title supplied by the editors that gave no hint of what it contained. Bush did not invent the word “hypermedia.” Ted Nelson would coin “hypertext” eighteen years later, in 1963. The Atlantic essay does not propose a computer; it proposes a desk. But every working hypermedia system that has existed since — Xanadu, NLS, HyperCard, NoteCards, Intermedia, Storyspace, Roam, every wiki, and the web — has cited Bush as the source of its founding claim. The claim was that the way human beings move through ideas is associative rather than hierarchical, that the machines we use for thinking should match the way we think, and that information becomes useful only when it can be linked to other information by the person who reads it. This chapter is about what Bush actually said, what he did not say, and how the essay landed in the men who built the systems that followed.

Who Bush was, and why anyone listened

Vannevar Bush was, by the summer of 1945, one of the most powerful figures in American science. He had been Dean of Engineering at MIT, then Vice President of MIT, then president of the Carnegie Institution in Washington. In 1940 he had persuaded Franklin Roosevelt to create the National Defense Research Committee, and in 1941 he had become Director of the Office of Scientific Research and Development, the agency that ran the American war effort in science and engineering. The Manhattan Project ran under Bush’s general oversight. The proximity fuze, radar at MIT’s Rad Lab, penicillin’s mass production, the first analog computers used in ballistics — these came out of organizations Bush either ran or had set up. When the war ended he was about to deliver a report to Roosevelt, Science, the Endless Frontier, that would propose the structure of postwar American science and lead, three years later, to the National Science Foundation.

“As We May Think” was not a side project. It was Bush thinking about what the scientists he had spent five years organizing were going to do after they stopped building weapons. He had a specific worry. The volume of scientific publication was growing past the point where any human could keep up with the literature of their field. Cross-disciplinary work, which the war had shown was where the breakthroughs came from, was made hopeless by the impossibility of reading across multiple specialties. The bibliographic infrastructure — card catalogs, indices, abstracts — was an accumulated mass of nineteenth-century technology that no longer fit the volume of twentieth-century knowledge. Bush thought this was the problem postwar science would have to solve before it could solve anything else. The essay is his answer.

The essay was first read by scientists, but it spread fast. Life magazine published an illustrated abridgment on September 10, 1945, with drawings by Alfred D. Crimi that visualized the device Bush had described. The Life version reached an audience orders of magnitude larger than The Atlantic’s, and it is the version that several of the people who later built hypermedia systems first encountered. Bush had a national platform. The essay was treated, in 1945, as a serious proposal from a serious figure. It is worth keeping this in mind, because the essay is sometimes presented now as a visionary speculation by a single thinker. It was that, but it was also a policy document by the man who had run American science, and it was read that way.

The Memex

The device Bush described he called the Memex. The name combined “memory” and “extender,” and was Bush’s coinage. The Memex was a desk. The desk had two slanted translucent screens at the user’s working height. Behind the screens was a microfilm storage system that could hold, in Bush’s estimate, what would now be called millions of pages. A keyboard let the user enter commands. A set of levers and pedals controlled the projection of microfilm pages onto the screens. A dry photography process let the user take notes that were merged back into the same microfilm store. The Memex was, in short, a workstation, designed before the word existed for a kind of work nobody had yet done.

Bush had not invented the Memex out of nothing. He had spent the 1930s working on what he called the Rapid Selector, a microfilm system that used photoelectric scanning to retrieve documents indexed by binary codes punched into the film. The Rapid Selector was a real machine. Versions of it had been built; one was used during the war for cryptographic indexing. The Memex was the Rapid Selector extended into a model of how a person might actually use such a machine for sustained intellectual work. The microfilm storage and the binary indexing were extrapolations of technology Bush had already built. The two screens, the dry photography, and the pedal-and-lever interface were extrapolations of contemporary office equipment. None of this was fantasy in the science-fictional sense. It was an engineering proposal, expressed at the level of mechanism, by an engineer who had been involved in building related machines for a decade.

Bush did not propose to build the Memex. He proposed it as a thing that could be built and that ought to be built. The essay’s claim is not “I have built this” but “this is what the next generation of scientists’ equipment should look like, and here are the reasons why.”

The reasons why — associative indexing

The reasons were specific. The bibliographic infrastructure Bush wanted to replace was hierarchical: a document had a place in a classification, a subject was filed under a heading, and to find a document you walked the hierarchy. This worked when the hierarchy fit the structure of human knowledge. It stopped working when the structure of human knowledge stopped being hierarchical, which Bush argued had already happened. The breakthroughs in science were being made at the joints between fields, and the bibliographic infrastructure had no good way to register joints. A paper on the structure of hemoglobin and a paper on cryptographic codebreaking might have a method in common — a kind of statistical inference, say — but the cataloging system would file them under different headings and the person who needed to see them together would have to know in advance that they were together.

Bush proposed that the right structure was associative rather than hierarchical. The human mind, he argued, works by association: one thought leads to another through a connection the thinker recognizes, not through a connection the cataloger has pre-registered. A machine designed to support human thinking should support that mode of operation directly. The Memex did so by letting the user create what Bush called “trails.” A trail was a sequence of documents that the user had linked together by their own act of linking. Once two documents had been put on a trail, a button on the Memex took you from one to the other in either direction. Multiple trails could pass through the same document. Trails could be named, saved, recalled, and — this is the crucial step — given to other people.

The trail is the founding object of hypermedia. It is not the document; it is the relation between documents, registered by a reader’s act of registering it. The Memex’s value was not the documents it stored — those were just microfilm — but the structure the user accumulated on top of them. Bush understood that this structure was the thing of value. He proposed that a scientist’s Memex, over the course of their career, would accumulate trails that constituted a record of how they had thought. The trails could be inherited. A senior scholar’s trails through the literature would be a kind of intellectual capital, transmissible to students and successors, more useful in many ways than the publications themselves because they would show not just what the scholar had concluded but how they had connected the conclusions to other people’s work.

This is the founding claim. It has two parts. The first is that the unit of intellectual value is the link, not the document. The second is that the production of links — the trail-making — is itself a creative act, by an author, that other people will want to read.

The trail-maker as a profession

Bush did not stop at saying that scientists would make trails for themselves. He proposed, explicitly, that a new profession would arise. He called its practitioners “trail blazers.” A trail blazer would be a person who made trails through the literature on behalf of others — who took a topic and built, in their own Memex, a structured path through what had been written on it, and then made that path available to anyone else who needed an entry into the topic. This was not librarianship in the existing sense. The librarian’s job was to file documents in a place where they could be found. The trail blazer’s job was to make sense of how documents related, in a form readers could traverse.

It is worth pausing on what this proposes. Bush was suggesting that the production of structured paths through other people’s writing would be a paying job, that the resulting paths would be objects in their own right, that they would be read in the same sense that articles are read, and that the trail blazer would be an author of the trail in the same sense that the article writers were authors of their articles. The structure was the work. The information was raw material.

There is no profession on the web that corresponds to this. The closest analogues are the curator of a personal blog who links extensively into a domain, the author of a long Wikipedia article whose footnotes are themselves a kind of trail, and the person who maintains a research bibliography in some specialty. None of these are paying professions at any scale. The link as a primary object of value, traded and consumed in the way the article is, did not survive into the world the web built. There are reasons for this and the chapters on the web’s technical decisions will treat them. Here it is enough to note that Bush’s proposal contained a labor market and the web’s world does not.

What the essay did not say

Several things commonly attributed to Bush are not in “As We May Think.”

Bush did not propose digital storage. The Memex was microfilm. The choice was not incidental; Bush had spent a decade on microfilm-based retrieval and considered it the appropriate storage technology for the scale and stability of scientific literature. A digital Memex was not imaginable from where he was sitting in 1945, and he did not imagine one.

Bush did not propose networking. The Memex was a personal workstation. The trails were transferable between Memexes — Bush described this — but the transfer was physical, by sharing microfilm. There is no proposal in the essay for a connected system of Memexes communicating over wires. The networked extension of the Memex idea was supplied later, by Engelbart and Nelson, and required them to combine Bush’s proposal with the very different proposal of packet-switched communication.

Bush did not propose a computer in anything like the modern sense. The Memex’s operations were mechanical and photographic. He did discuss, in the same essay, the possibility of more powerful computing machinery — he was aware of the calculators that had been built during the war — but the Memex itself was not described as a computer. It was an information-retrieval device.

Bush did not propose direct two-way linking in the modern hypertext sense. The trails were sequences; you walked them in order. Branching trails are described but the system is closer to a linear path with named entry and exit points than to the dense web of bidirectional citations that Xanadu would later propose. The richness Nelson would put into “hypertext” was, in Bush, present in seed form but not in the elaborated form.

Bush did not propose the elimination of hierarchy. The Memex coexisted, in his description, with the existing hierarchical bibliographic infrastructure. Trails were a supplement to classification, not a replacement for it. The opposition between hierarchy and association that became the founding tension in the hypermedia literature is something later writers read back into Bush; the essay itself is ecumenical.

These are not faults of the essay. They are limits of its proposal, and they matter because the next forty years of hypermedia work consisted, in large part, of extending Bush’s proposal in each of these directions — into digital storage, into networking, into computation, into richer link structures, into a serious account of how associative structure might displace hierarchical organization. Each extension was a step away from Bush and toward something more ambitious. Each was also a step toward systems that turned out to be harder to build than their proposers thought.

How the essay landed

The essay was read by, among others, two men who would do more than anyone else to turn its proposal into working machinery.

Douglas Engelbart read either The Atlantic version or, more likely, the Life abridgment in late 1945. He was a Navy radar technician stationed in the Philippines, waiting for redeployment, and the story he told later was that he found the issue in a Red Cross library on Leyte and was struck by the Memex’s claim that machines could be built to support thinking rather than merely to compute. The exact issue and exact location have been told differently in different interviews, but the substance is consistent: Engelbart, in his early twenties, read Bush, took the proposal as a charge, and over the next two decades built the system — NLS, treated in chapter six — that was the first working version of what Bush had imagined.

Ted Nelson read Bush later, as a graduate student. By Nelson’s own account in Literary Machines and elsewhere, the encounter with “As We May Think” was formative. He was already thinking about nonlinear writing as a literary form; Bush gave him the engineering vocabulary to think about it as a system. The word “hypertext,” coined by Nelson in 1963 and first published in 1965, is an extension of Bush’s trail in the direction of writing as such. Where Bush imagined the trail-maker as a kind of trained scholar, Nelson imagined the hypertext as the natural form of writing itself, with every author producing nonlinear documents and every reader navigating them through links the author had built and links the reader chose.

The hypermedia academics at Brown — Andries van Dam, Nicole Yankelovich, Norman Meyrowitz — cited Bush as the origin of their tradition, and built systems (HES, FRESS, Intermedia) that were closer to Bush’s specific proposal in their treatment of trails as named, navigable, transmissible objects. So did the team at Xerox PARC who built NoteCards. So did the literary hypertext community. Each system extended Bush in different directions, but each acknowledged the debt, and each could be seen as one possible answer to the question Bush had raised.

The web, for what it is worth, also cites Bush. Tim Berners-Lee’s original 1989 CERN proposal opens with a discussion of the difficulty of cross-referencing information across an organization and notes Bush’s earlier work in the lineage of the problem. The lineage is acknowledged. The proposal is, however, much less than Bush’s. Berners-Lee’s web does not have trails. It does not have named navigable sequences. It does not have the proposition that the link is a first-class object that an author produces and a reader consumes. It has documents and one-way pointers between them, and the pointers are not registered with the destinations and do not persist if the destination moves. The web is, by Bush’s standards, a half-implementation of the founding proposal — the documents are there but the trails are not.

The shape of the surviving claim

What survives, across every hypermedia system from 1945 to the present, is the proposition that the structure connecting documents is itself a document, that this meta-document is the thing of intellectual value, and that machines should be designed to make the production and consumption of that meta-document easy. The systems differed enormously in how they realized this proposition. Bush’s trails were linear and microfilm-based; Engelbart’s were structured and outline-based; Nelson’s were branching and parallel; the academic hypermedia systems made them composable with formal argument structures; HyperCard made them clickable buttons on stacked cards; the literary hypertext canon turned them into the substance of narrative. But the claim is the same. Whatever the link is, it matters more than the documents.

The web’s version of the claim is degenerate. A web link is a one-way pointer from one document to another, registered nowhere, persisted by nothing except convention, and produced as a side effect of writing the source document. It has no name. It cannot be inherited. It cannot be reviewed independently of the source. It can be broken silently by anyone other than the author who made it. It carries no semantics — there is no way to say “this is a citation,” “this is a refutation,” “this is a quotation.” There is, in the bare web, no analogue of Bush’s trail. There is an <a href> tag.

This loss is the central technical loss the book is about, and it is the source of everything else. The application has eaten the document; the document was supposed to eat the link; the link was supposed to be the author’s primary contribution to the shared infrastructure of thought. None of that happened. We have applications now, and the documents inside them are increasingly disposable, and the links between the documents — when there are any — are managed by the applications rather than by the authors. This is not what Bush proposed. It is not what any of the systems descended from Bush proposed. It is what we have, and the rest of this book is an attempt to understand how we got here.

The next chapter turns from hypermedia to the network underneath it. The founding claim of the internet is older than the founding claim of hypermedia, and it is differently structured. Where Bush imagined a single workstation with extraordinary capacity, the people who built the internet imagined a federated mesh of peers, each independent, each contributing to a shared substrate that no one owned. That vision, like Bush’s, was held in clear view for two decades before it began to be eroded. Understanding it is the second half of the frame.